Introduction
Abnormal affective biases have been reported in bipolar disorder (BD), including among euthymic and clinically stable patient groups. In the majority of studies, the observed emotional bias appears to be negative in nature (Gopin et al. Reference Gopin, Burdick, DeRosse, Goldberg and Malhotra2011; Lemaire et al. Reference Lemaire, Aguillon-Hernandez, Bonnet-Brilhault, Martineau and El-Hage2014). However, observed patterns of emotional bias are inconsistent, and differences with healthy controls are not always observed (Venn et al. Reference Venn, Gray, Montagne, Murray, Burt, Frigerio, Perrett and Young2004; Rock et al. Reference Rock, Goodwin and Harmer2010). Understanding the emotional biases present in BD is important, as they are posited to underpin some of the symptoms of mood disorders as well providing a potential target for therapeutic intervention (Gotlib & Krasnoperova, Reference Gotlib and Krasnoperova1998; Harmer et al. Reference Harmer, Goodwin and Cowen2009).
Mood instability is regarded as clinically important – if difficult to define – and characterized in part by frequent shifts in affective category, but also incorporating features such as abnormal intensity in affect and exaggerated responses to psychosocial cues (Koenigsberg, Reference Koenigsberg2010). Mood instability is a widely recognized feature in many psychiatric disorders, including BD (Henry et al. Reference Henry, Van den Bulke, Bellivier, Roy, Swendsen, M'Bailara, Siever and Leboyer2008). Greater mood instability in BD, as indexed by the proportion of weeks featuring a shift between euthymia, clinical, and subclinical mood states, has been associated with functional impairment (Strejilevich et al. Reference Strejilevich, Martino, Murru, Teitelbaum, Fassi, Marengo, Igoa and Colom2013) pointing to the clinical relevance of inter-episode sub-syndromal mood patterns. Accordingly, successful management of BD is thought to necessitate the maintenance of mood stability (Malhi et al. Reference Malhi, McAulay, Das and Fritz2015). Furthermore, a recent study using daily mood monitoring with 27 euthymic BD patients and healthy controls demonstrated that greater mood instability [measured as mean square successive differences (MSSD) of mood ratings] was correlated with greater functional impairment and was predictive of depressive symptoms (Gershon & Eidelman, Reference Gershon and Eidelman2015). However, nothing is known of how aspects of cognition, including emotional processing, might be associated with fluctuations in mood symptoms.
Inconsistent reports of emotional bias in BD is likely due to a number of factors including varying inclusion criteria and measures of emotional processing, but also small sample sizes and poor classification of current mood. In addition, several studies have reported state-dependent attentional biases in BD (Murphy et al. Reference Murphy, Sahakian, Rubinsztein, Michael, Rogers, Robbins and Paykel1999; Garcia-Blanco et al. Reference Garcia-Blanco, Perea and Livianos2013). The effects of medication are a further complicating factor, and such effects may be substantial: for example, Holmes and colleagues found that medicated subjects with BD (currently taking lithium or sodium valproate) were both less accurate and slower to respond to happy words compared to unmedicated subjects with BD in an affective shift task (Holmes et al. Reference Holmes, Erickson, Luckenbaugh, Drevets, Bain, Cannon, Snow, Sahakian, Manji and Zarate2008). However, many studies have been underpowered to explore the potential effects of medication in patient groups, which are argued to not account for core observations (Lex et al. Reference Lex, Meyer, Marquart and Thau2008; Garcia-Blanco et al. Reference Garcia-Blanco, Perea and Livianos2013; Lemaire et al. Reference Lemaire, Aguillon-Hernandez, Bonnet-Brilhault, Martineau and El-Hage2014). Inconsistencies in previous literature may also be driven by possible differences in emotional processing between BDI and BDII, which few studies have been powered to explore (Mercer & Becerra, Reference Mercer and Becerra2013).
Here, we explored associations between emotional bias, as measured in an emotional categorization and memory task, and prospectively gathered weekly manic and depressive symptoms, in a large, naturalistic cohort of BD patients. Given the evidence linking mood instability with greater impairment in BD, we hypothesized that higher levels of symptom lability would be associated with a well-validated risk factor for affective disorders: greater bias towards negative stimuli. The size of our cohort allowed us to explore the effects on cognition of core demographic and clinical variables (including current mood symptoms, current medication status, and effects of diagnostic subtype), and to determine whether instability in symptoms of mood episodes added anything over and above these more established predictors.
Method
Participants
Participants were recruited into the OXTEXT-1 cohort study; recruitment began in 2010 and is ongoing. Here we report data of all participants who completed OXTEXT-1 assessments between 4 May 2010 and 18 June 2014. OXTEXT-1 was approved by Oxfordshire Research Ethics Committee A and written informed consent was obtained from all participants (REC ref. 10/H0604/13).
Recruitment into OXTEXT-1 is via three main routes: by referral from an Adult Mental Health Team within Oxford Health NHS Foundation Trust; from posters and fliers that are displayed in the local community; and by clinicians at the Bipolar Disorder Research Clinic at the NIHR-Clinical Research Facility (NIHR-CRF) at the Warneford Hospital (Oxford, UK). Participants were eligible to take part in the study if: they had a primary diagnosis of DSM-IV-R bipolar disorder [BDI, BDII and BD ‘not otherwise specified’ (NOS)]; were aged ⩾16 years; and both willing and able to give informed consent to participate.
Procedure
Research assessment
All participants completed a single baseline assessment. During this assessment, participants completed an audio-recorded diagnostic interview, adapted from the Mini International Neuropsychiatric Interview (MINI; Sheehan et al. Reference Sheehan, Lecrubier, Sheehan, Amorim, Janavs, Weiller, Hergueta, Baker and Dunbar1998), and conducted by a trained research assistant. All clinical interviews were reviewed by a research psychiatrist for diagnostic confirmation. A baseline research questionnaire was used to collect demographic and clinical information including current medications. Participants then completed the Emotional Categorization and Memory Test as part of a battery of neuropsychological tasks. Data relating to other tasks are not presented here. All participants were then registered on the True Colours self-monitoring system (see Mood Assessment section, below) and provided training on how to use the system.
Emotional Categorization and Memory Test
The ‘Emotional Categorization and Memory’ Test utilizes positively- and negatively-valenced word stimuli to explore emotional biases in processing. It is comprised of three separate tasks and has been previously described (see Harmer et al. Reference Harmer, Heinzen, O'Sullivan, Ayres and Cowen2008). A brief description of the tasks is included below.
Emotional Categorization Task
Participants were presented with 60 positively- or negatively-valenced personality characteristics chosen to be agreeable or disagreeable (e.g. cheerful, generous v. domineering, hostile; Anderson, Reference Anderson1968). Each word was presented in the centre of a laptop screen for a duration of 500 ms. Participants were asked to indicate whether they would like or dislike being described as possessing each personality characteristic by pressing one of two keys on the keyboard, as quickly and as accurately as possible.
Emotional Memory Task
Following an approximately 15-min delay, participants were asked to freely recall (and write down), within a two minute time window, as many of the personality trait words from the above Emotional Categorization Task as possible.
Emotional Recognition Task
Finally, 120 personality characteristic words were presented in the centre of the laptop screen, one at a time and for duration of 500 ms. These words consisted of 60 familiar words presented in the first Emotional Categorization Task (30 agreeable, 30 disagreeable) and 60 novel personality characteristic words (30 agreeable, 30 disagreeable and again taken from Anderson, Reference Anderson1968). Participants were asked to categorize each of the personality traits as either ‘familiar’ from the first task, or ‘novel’, and respond by pressing one of two keys on the keyboard as quickly and accurately as possible. Target sensitivities:
$$\delta^{\prime} = 0.5 + [(y - x).(1 + y - x))/(4y(1 - x)],$$
where y is the probability of correct responses and x is the probability of false alarms, as well as response biases:
$$\beta = (y(1 - y) - x(1 - x))/(y(1 - y) + x(1 - x))$$
were calculated. Higher sensitivity has been shown to correlate with higher accuracy in the task, while response bias allows filtering of data to remove cases where participants are notably over-favouring ‘familiar’ or ‘novel’ responses throughout the task (Mocking et al. Reference Mocking, Pflanz, Pringle, Parsons, McTavish, Cowen and Harmer2013).
Mood assessment
All participants were registered to submit mood ratings on a weekly basis by answering text or email prompts from the True Colours self-monitoring system (Miklowitz et al. Reference Miklowitz, Price, Holmes, Rendell, Bell, Budge, Christensen, Wallace, Simon, Armstrong, McPeake, Goodwin and Geddes2012). Depression ratings were captured with the Quick Inventory of Depressive Symptomology (QIDS; Rush et al. Reference Rush, Trivedi, Ibrahim, Carmody, Arnow, Klein, Markowitz, Ninan, Kornstein, Manber, Thase, Kocsis and Keller2003) and mood elevation ratings with the Altman Self-Rating Mania scale (ASRM; Altman et al. Reference Altman, Hedeker, Peterson and Davis1997).
Indices of manic and depressive symptoms were extracted from the True Colours system both for the week in which participants undertook the neuropsychological assessment (week 0; denoted W0), and for the proceeding 6 weeks (i.e. W0–W6). Thus all self-reported measures of depression and mania were collected in the 6- to 7-week period following neuropsychological assessment.
For the purpose of data analysis the ASRM data obtained from True Colours responses were transformed into binary data (⩽5 and >5) as data was not normally distributed. ASRM scores >5 are considered to indicate a manic state, while those ⩽5 are considered to indicate a non-manic state (Altman et al. Reference Altman, Hedeker, Peterson and Davis1997).
Mood instability
We used the square root of the mean square successive difference (MSSD) of weekly QIDS and ASRM scores as a measure of depressive and manic symptom instability. MSSD measures provide an index of data dispersion while accounting for variability and the temporal (serial) aspects of data, and has previously been used as a measure for affective instability in BD (Gershon & Eidelman, Reference Gershon and Eidelman2015) and borderline personality disorder (BPD) (Jahng et al. Reference Jahng, Wood and Trull2008). Square-rooting of MSSD scores (denoted here as RMSSD) normalizes the inherent positive skew of MSSD data for parametric analyses.
Statistical analysis
All statistical analyses were conducted in SPSS (IBM SPSS Statistics v. 22, IBM Corp., USA). χ² tests were used to test for differences between diagnostic groups (BDI v. BDII) in terms of gender, education level (categorized as less than GCSE; O-level/GCSE; A-level or equivalent; Degree/NVQ level 5; and postgraduate degree), current medication (with lithium, antidepressant, antipsychotic, mood stabilizer, or drug free coded as separate binary variables), ethnicity, and baseline evidence for mania (ASRM > 5). Diagnostic group differences for continuous variables (e.g. age, baseline depression, RMSSD values) were tested using independent-samples t tests.
Base model
A base backwards linear regression model was used to test the effects of diagnosis, clinical variables, and mood variables on emotional processing. This base model contained the following independent variables: gender; diagnosis; age; education level; baseline mood (QIDS score and categorical ASRM ⩽5/>5 at W0); current medications or drug free status (coded as dummy variables); and QIDS RMSSD or ASRM RMSSD (entered separately). Dependent variables were selected as described below.
Separate linear regression analyses were conducted for each of the following dependent variables. To examine which demographic and clinical factors impacted overall task performance, we entered into the base model (individually) dependent variables of: (i) categorization accuracy; (ii) categorization reaction time; (iii) accurate free recall; (iv) recognition accuracy; and (v) sensitivity in the recognition task. In all cases dependent variables collapsed data across positively- and negatively-valenced trials. We further tested how demographic and clinical variables were associated with emotional bias, by selecting as dependent variables (again individually), (vi) bias in categorization accuracy (positive – negative, such that a positive value reflects an overall positive emotional bias); (vii) bias in accurate recall, (viii) bias in sensitivity (recognition task), and (ix) bias in categorization reaction time (RT).
We also examined whether depressive and manic/hypomanic symptom instability was associated with demographic and clinical variables (e.g. age, diagnostic subtype, current medication). To do this, we re-ran the main linear regression model but with QIDS RMSSD and ASRM RMSSD selected as dependent variables rather than independent variables.
Results
The data of 346 participants were extracted. The total number of participants diagnosed with BD-NOS or those with unconfirmed diagnoses was low (N = 24, 6.9%). Early analyses suggested that data collected from the BD-NOS group was heterogeneous, such that making conclusions based on such a small sample would be challenging. Given that we wished to test for possible effects of the more prevalent BDI and BDII diagnostic subtypes on emotional processing, these participants were therefore excluded from further analysis. Due to procedural changes in the study protocol medication data was not collected within 30 days of testing for a subset of 51 participants (15.0%) which were excluded from further analysis (remaining N = 271). For the results reported, each regression analysis included all available data for the variables included in the respective models. Models including RMSSD variables included only those participants who responded at least four times to weekly questionnaires during the 7-week period from baseline (W0) to W6, leading to the exclusion of 26 participants’ data (9.5%) and thus a final sample for the RMSSD analyses of N = 245. The groups excluded and included on the basis of response rate are well matched for age, gender and diagnosis (see Supplementary Table S1).
Participant demographics and clinical variables
Demographics and clinical characteristics of the sample is shown in Table 1. As the sample was predominantly Caucasian (94%) ethnicity is presented and analysed as a binary variable (Caucasian v. non-Caucasian). Means ± standard errors (s.e.) are provided in the text unless otherwise indicated. Unstandardized beta coefficients are presented with s.e. in parentheses.
Demographic and clinical characteristics of participants
QIDS, Quick Inventory of Depressive Symptomology; ASRM, Altman Self-Rated Mania scale.
Values are numbers of cases (percentages in parentheses) unless otherwise specified.
a Educational information missing for three BDI participants and two BDII participants. Percentages are reported with regards to the data available.
A full comparison of the diagnostic groups (BDI v. BDII) on the clinical and demographic variables presented in Table 1 is provided in the Supplementary material. Diagnostic groups differed in terms of prevalence of lithium treatment (χ2 1 = 5.386, p = 0.020), with 38% of BDI patients v. only 24% of BDII patients on lithium. They also differed on the prevalence of antidepressant treatment (χ2 1 = 4.357, p = 0.037), with 32% of BDI patients v. 45% of BDII patients on antidepressants. However, the groups did not differ in terms of prevalence of anticonvulsant treatment (χ2 1 = 0.024, p = 0.877), antipsychotic treatment (χ2 1 = 0.699, p = 0.403), or drug-free status (χ2 1 = 0.728, p = 0.393).
Diagnostic groups showed similar values of RMSSD for depressive symptoms at W0–W6 (t 243 = −0.529, p = 0.597) and RMSSD for manic/hypomanic symptoms at W0–W6 (t 243 = −1.122, p = 0.263) (see Table 2).
RMSSD values, as a measure of mood instability (N = 245)
RMSSD, Root mean square successive difference; QIDS, Quick Inventory of Depressive Symptomology; ASRM, Altman Self-Rated Mania scale.
Demographic and clinical predictors of depressive/manic symptom instability
Current antipsychotic use was associated with decreased instability in manic symptoms (2.75 ± 0.20 compared to 3.30 ± 0.25 for those not taking antipsychotics [B = −0.780 (0.343), p = 0.024]). As expected, baseline symptomology was significantly and positively related to RMSSD scores. Evidence of mania at baseline (i.e. W0 ASRM scores >5) was associated with significantly higher ASRM RMSSD values [4.79 ± 0.45 v. 2.57 ± 0.18 (B = 2.280 (0.402), p < 0.001)]. Higher levels of baseline depressive symptoms (W0 QIDS scores) were also related to higher QIDS RMSSD values [B = 0.088 (0.031), p = 0.004]. In addition, increased age was linked with lower instability in manic symptoms [B = −0.035 (0.013), p = 0.008]. Finally, female participants showed higher levels of instability in depressive symptoms compared to male participants [4.06 ± 0.21 v. 3.39 ± 0.28 (B = 0.758 (0.364), p = 0.039)].
Predictors of overall task performance
Effects of demographic variables (including age, gender, and education) on overall performance on the emotional categorization and memory tasks are presented in the Supplementary material. In brief, participants with higher education levels tended to perform both more quickly and accurately in most tasks, while increasing age was associated with poorer performance and slower responses. Females tended to be quicker and more accurate in the majority of tasks compared to males.
Effects of diagnosis
Participants diagnosed with BDII were able to accurately recall more words than BDI participants, but this effect did not reach significance (3.49 ± 0.19 v. 3.07 ± 0.14, respectively) [B = 0.368 (0.208), p = 0.077]. Participants with BDII were significantly quicker in the recognition task than participants with BDI [1554.53 ± 39.16 v. 1756.57 ± 48.20 (ms), respectively] [B = −163.771 (68.313), p = 0.018].
Effects of medication
Participants who were currently drug free demonstrated overall better sensitivity in the recognition task (0.891 ± 0.048 v. 0.840 ± 0.75) [B = 0.030 (0.014), p = 0.037] while those on anticonvulsants demonstrated poorer sensitivity than those not on anticonvulsants (0.830 ± 0.086 v. 0.856 ± 0.063) [B = −0.020 (0.009), p = 0.021] and those currently on lithium demonstrating longer reaction times in the recognition task than those not currently taking lithium [1805.82 ± 68.30 (ms) v. 1596.79 ± 32.24] [B = 157.155 (70.217), p = 0.026].
No effects of baseline mood or mood instability were observed on behavioural outcomes related to overall task performance.
Predictors of emotional bias
Effects of diagnosis
No effects of diagnosis on emotional bias (as indexed by positive v. negative categorization accuracy, free recall, recognition accuracy, or categorization reaction time) were observed.
Effects of medication
Those participants who were drug free were quicker to categorize negative words relative to positive words as compared with participants on any psychotropic medication (mean difference in categorization reaction time −62.09 ± 8.91 for participants on any medication, −18.94 ± 30.05 for medication-free) [B = 85.101 (32.951), p = 0.011] (see also Supplementary Table S2).
Effects of baseline mood symptoms
Participants scoring >5 on the ARSM at baseline demonstrated greater accuracy for categorization of positive words relative to negative words (0.62 ± 0.64 v. −0.55 ± 0.23) [B = 1.326 (0.518), p = 0.011]. Participants with a higher baseline QIDS score were faster to categorize negative words relative to positive words in the categorization task [B = 3.445 (1.525), p = 0.025].
Effects of mood instability on emotional bias
Greater categorization accuracy for negative words relative to positive words was related to higher instability in depressive symptoms [B = −0.204 (0.092), p = 0.027] (Fig. 1). In the recall task, greater relative recall of negative words was related to higher instability in manic symptoms [B = −0.124 (0.055), p = 0.026] (Fig. 2). No other associations between symptom instability and emotional bias were observed.
Scatterplot, with a line of best fit, showing the mean difference in categorization accuracy (positive – negative; higher values are indicative of greater positive bias) against Quick Inventory of Depressive Symptomology-root mean square successive difference (QIDS RMSSD). The relationship between bias in categorization accuracy and QIDS RMSSD remains significant after the removal of a potential outlier [filtering out difference in categorization accuracy >20 (B = −0.163 (0.081), p = 0.045)].
Scatterplot, with a line of best fit, showing the mean difference in free recall (positive – negative; more positive values are indicative of greater positive bias) against Altman Self-Rated Mania scale-root mean square successive difference (ASRM RMSSD).
Exploration of R 2 values determined that the explained variance in categorization accuracy of negative v. positive words, including instability in depressive symptoms (QIDS RMSSD) as a regressor, was 16.9%. Removing QIDS RMSSD from the model caused a drop in variance explained by 2.3% (to 14.6%). The total variance explained in recall of positive relative to negative words was 3.2%, which reduced by 1.7% when instability in manic symptoms was removed from the model.
Discussion
We observed that greater instability for depressive symptoms and manic symptoms correlated with negative biases in categorization and memory, respectively. In addition, we observed an association between greater positive bias (faster responses in categorizing positive compared to negative words) and current medication use as compared with non-use, as well as mood-congruent effects on emotional processing, such that higher levels of current depressive symptomology were related to greater negative bias (faster categorization of positive relative to negative words) while self-reported levels of substantive manic symptoms were related to greater positive bias (more accurate categorization of positive relative to negative words). Additionally, current antipsychotic use was associated with reduced instability in manic symptoms. Notably, we find no evidence that a diagnosis of BDI or BDII mediates emotional bias, although some evidence of more general cognitive impairment in BDI was found.
Our results are the first to draw links between emotional biases and instability in both depressive and manic symptoms in BD. The degree of variance in behavioural data explained by self-reported symptom instability was modest. Nevertheless, our result thus suggest that mood instability in BD may be related to underlying, and potentially clinically relevant, patterns of emotional information processing. That frequent shifts in mood state were related to maladaptive patterns of information processing might be understood in terms of cognitive reactivity, whereby changes in affect – including sub-syndromal changes – may reactivate cognitive patterns that were present during past mood episodes (Lau et al. Reference Lau, Segal and Williams2004). Here, instability in both manic and depressive symptoms were linked to increased negative emotional bias, evidence that fluctuations in the symptoms of either affective polarity are clinically meaningful. It is possible that the shifting of mood in and of itself is important in underpinning the pathology of BD, as well as the experience of mood episodes. More frequent or exaggerated mood shifts, whether towards the manic or depressive pole, may be associated with de-stabilized cognitive schemas. Effective coping strategies may also need to be more adaptive and dynamic, and more challenging therefore to develop or maintain. Variability in mood, uncertainty for the individual in predicting mood changes, and destabilization of cognitive schemas and coping strategies may lead to weighting of information processing towards more negatively-valenced cognitive stances, in a disorder where depressive symptoms are known to dominate the longitudinal course, and where depressive symptoms and episodes may be more sailent.
In contrast, current mood state appeared to have differential effects on emotional processing, with current depressive and manic symptoms associated with greater negative and positive bias, respectively (as observed also by others; Murphy et al. Reference Murphy, Sahakian, Rubinsztein, Michael, Rogers, Robbins and Paykel1999; Garcia-Blanco et al. Reference Garcia-Blanco, Perea and Livianos2013). That mood instability is more parsimoniously linked to negative cognitive bias may highlight the importance of affective lability in understanding the pathology of BD as well as promote it as a potential target in experimental models of treatment efficacy.
Our findings also add to a body of work suggesting that medications used in the treatment of mood disorders may exert their therapeutic effects via the modification of emotional information processing patterns. Current use of medication as compared with non-use in our bipolar cohort was associated with more positive cognitive bias, as measured by faster categorization of positive relative to negative word stimuli. Notably, in one of the few other studies to explore emotional biases in a community sample of depressed patients, current use of any medication (compared to non-use) was associated with enhanced processing of positive stimuli in patients with major depressive disorder (Wells et al. Reference Wells, Clerkin, Ellis and Beevers2014). The current findings are therefore consistent with the hypothesis that medication used in bipolar illness may also enhance the processing of positive cues, though this needs to be examined in a controlled experimental design to rule out the influence of possible confounders.
Our data also point to the potential effects of antipsychotic medication in reducing lability in manic symptoms amongst our patient participants. This observation is consistent with previous clinical research linking antipsychotic use with the amelioration of manic symptoms in bipolar disorder (Vieta & Goikolea, Reference Vieta and Goikolea2005). In addition, a recent review has suggested that olanzapine and aripiprazole significantly reduce mood instability in borderline personality disorder (Bellino et al. Reference Bellino, Bozzatello, Brignolo and Bogetto2012). Although establishing the effects of medications necessarily involves carefully controlled experimental work, our data collected from a naturalistically recruited cohort of BD patients highlight the potential of antipsychotic treatments to stabilize manic symptoms in BD over time, rather than solely reduce symptoms at a given time point: greatest clinical benefit is likely to involve both.
We observed that BDI as compared to BDII was associated with relatively impaired free recall, and slower responses in word recognition, but diagnostic groups did not differ in terms of emotional processing outcomes. This is broadly consistent with previous literature, which has focused on predominantly non-affective cognitive function, and points to greater levels of cognitive impairment in BDI compared to BDII (Simonsen et al. Reference Simonsen, Sundet, Vaskinn, Birkenaes, Engh, Hansen, Jónsdóttir, Ringen, Opjordsmoen, Friis and Andreassen2008; Hsiao et al. Reference Hsiao, Wu, Wu, Hsu, Chen, Lee, Lee, Yeh, Yang, Ko and Lu2009; Bourne et al. Reference Bourne, Bilderbeck, Drennan, Atkinson, Price, Geddes and Goodwin2015). Of the few other studies comparing BDI and BDII groups on affective cognitive function, results seem to be inconsistent (Mercer & Becerra, Reference Mercer and Becerra2013). For example, one previous study found that euthymic BDI participants were worse at labelling some emotional facial expressions compared to BDII participants (Derntl et al. Reference Derntl, Seidel, Kryspin-Exner, Hasmann and Dobmeier2009) while another found no differences between groups (Lembke & Ketter, Reference Lembke and Ketter2002). Our results lend support to the idea that emotional processing in BDI and BDII do not differ substantially, based on an analysis of a large cohort in which we could concurrently explore the effects of both current mood symptoms and medications on cognitive performance.
Limitations
This is the first study to explore the association between emotional bias and symptom instability in BD. However, the conclusions that can be drawn on the basis of these data are subject to a number of important caveats. First, we had no data from a matched healthy control group, without which it is unclear what patterns of emotional bias and mood instability might be within the normal range. As this is the first study – to the best of our knowledge – which reports emotional processing in BD as measured in the emotional categorization and memory task, our approach was exploratory. We did not correct for multiple comparisons, and thus our findings require replication. Our data suggest that emotional biases may be predictive of symptom variability, but establishing this causal relationship is beyond the scope of this correlational study; we also lacked mood data in the weeks prior to task completion, which limited the extent to which we could explore the temporal relationship between mood and emotional bias. The putative effects of current medication on emotional bias may be driven by unknown clinical variables which differentiate those patients who receive pharmacotherapy from those who do not, although there is no evidence to suggest that patients who are concordant with medication are more likely that those who are not to have a pre-existing positive cognitive bias.
While medication information was available for this sample, and was included in our analyses – unlike many other studies (Lex et al. Reference Lex, Meyer, Marquart and Thau2008; Gopin et al. Reference Gopin, Burdick, DeRosse, Goldberg and Malhotra2011; Gershon & Eidelman, Reference Gershon and Eidelman2015) – we did not collect information about dosage. We did not screen for co-morbid Axis I or II disorders, raising the possibility that some of our findings are driven by symptom patterns associated with personality disorders. In particular, traits of BPD, characterized itself in part by lability of mood, and frequently co-morbid with BD (Magill, Reference Magill2004) might have been expected to have influenced our findings.
Reliance on patient- reported mood symptoms may have introduced bias into our data, although previous research shows promising agreement between self-report and clinician-reported longitudinal symptoms in BD (Born et al. Reference Born, Amann, Grunze, Post and Scharer2014). The sensitivity of our symptom-tracking approach to detect mood fluctuations is still unclear, but was pragmatic in enabling un-intrusive data collection on a larger scale, and possibly with higher compliance, than might otherwise be feasible. Finally, the best methods of characterizing and quantifying mood instability, including frequency and length of overall time-frame of mood sampling, remain uncertain. Although we used a measure of mood instability which been employed in previous research (Jahng et al. Reference Jahng, Wood and Trull2008) other measures of signal variability exist (see Garrett et al. Reference Garrett, Samanez-Larkin, MacDonald, Lindenberger, McIntosh and Grady2013) and may prove to be more sensitive, but typically require more data-points than were available in the present study. Our mood episode data was gathered on a weekly basis, although fluctuations on finer time-scales are likely to be clinically relevant. New mobile technologies are making frequent sampling of mood increasingly practicable (Faurholt-Jepsen et al. Reference Faurholt-Jepsen, Vinberg, Christensen, Frost, Bardram and Kessing2013).
In conclusion, our data suggest that maladaptive emotional biases may be associated with greater mood variability in BD. The observed links between medication use, emotional bias, and mood patterns point to the clinical relevance of emotional biases in BD, as well as the potential of both emotional bias and quantified mood instability to act as potential therapeutic targets or markers of treatment response.
Supplementary material
For supplementary material accompanying this paper visit http://dx.doi.org/10.1017/S003329171600180X.
Acknowledgements
This paper presents independent research funded by the National Institute for Health Research (NIHR) under its Programme Grants for Applied Research Programme (Reference Number RP-PG-0108-10087). G.M.G. and J.G. are NIHR Senior Investigators. We also acknowledge the provision of facilities for conducting the study by the NIHR Oxford Cognitive Health and Clinical Research Facility, Warneford Hospital, Oxford. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health. A.C.B. and L.A. are supported by a Wellcome Trust Strategic Award (CONBRIO: Collaborative Oxford Network for Bipolar Research to Improve Outcomes, Reference number 102616/Z).
Declaration of Interest
G.M.G. reports grants and personal fees from Servier, personal fees from Teva, personal fees from Otsuka, personal fees from Takeda, grants and personal fees from Lundbeck, personal fees from Ei Lilly, personal fees from Merck, personal fees from GSK, personal fees from Astra Zeneca, and grants from P1vital during the conduct of the study. C.J.H. has received consultancy fees from Lundbeck, Servier and P1vital. She has shares in P1vital and also is a company director and shareholder of Oxford Psychologists Ltd. C.J.H. has received grant funding from J&J, UCB, Lundbeck, Sunovion, AstraZeneca, Servier and GSK. All other authors declare no conflicts of interest.
